Do you still remember the time when ray tracing on mobile was just a buzzword and nothing more? Arm hopes to make you erase all of that from your memory, as the newly-launched Mali G2-Ultra NX graphics processing unit is definitely the biggest gamble in trying to render your mobile phone graphics as if it were an actual gaming computer. And guess what technology made it happen.
Also read: Game Development, AI, VFX: Why Indian colleges need gaming PCs
The Mali G2-Ultra NX from Arm is the first “AI-native” GPU, which basically translates into having neural accelerators incorporated directly into shader cores instead of being just bolted on later on. That is a rather important architectural decision, since it allows neural computations to utilize the same memory hierarchy and cache as graphics and compute. And that reduces overhead due to data transfer between different processing engines.
And on top of that, Arm calls this their biggest Mali instruction set enhancement in seven generations, offering up to 2x increase in the number of registers per warp to handle complex scenes. Here are some of the numbers provided by Arm: up to 24% higher performance of the benchmarks suite, up to 14% boost to the non-AI gaming, and the third generation of ray tracing core which reduces DRAM bandwidth consumption by up to 13% for ray tracing benchmarks. Also, support for the Opacity Micromaps feature is mentioned, which in one Arm demo increased the frames per second by 30%, reducing the ray tracing workload by 70%.
Those numbers come from Arm’s own demos, not independent benchmarks, and Mali G2-Ultra NX itself is IP rather than a chip you can walk into a store and buy. Silicon partners like MediaTek and Samsung still need to license it and bake it into actual chipsets before any of this shows up in a phone near you.
Also read: Ceiling fan and air purifier: Does better air circulation mean faster cleaning?
Those familiar with recent gaming laptops should have no trouble seeing that the similarities exist. Mali G2-Ultra NX comes equipped with three neural graphics technologies that look like an almost direct translation of DLSS technology to phones.
Neural Super Sampling (NSS) creates a higher-resolution image from a lower-resolution rendering through a neural network, just as DLSS Super Resolution does. Neural Frame Rate Upscaling creates entirely new intermediate frames using motion and depth information, similarly to what DLSS Frame Generation does. Neural Super Sampling and Denoising, which upscales using neural networks for ray tracing scenes while reducing noise levels.
However, there is one major distinction: philosophy. While Nvidia traditionally keeps DLSS within its closed system of technologies and products, Arm is doing the opposite by developing an open software stack that includes Vulkan machine learning extensions and Unreal Engine plugins in a manner similar to how AMD develops its FSR technology. Moreover, Arm claims it created Neural Dawn, a demonstration of the technology in cooperation with Sumo Digital, that can show up to 4x improvements in efficiency and 70 percent reduction in memory usage.
Here’s where things get really exciting for gamers. There hasn’t been a time when mobile chipsets haven’t struggled with the trade-off between performance, battery efficiency, and overheating. However, while neural graphics does not solve that problem, it changes the way the fight will play out by reducing the need for actual rendering.
If everything goes according to plan, the result will be smartphones able to sustain console-grade graphics running up to 120 frames per second without turning into hand warmers. Developers are already working with Where Winds Meet using Arm’s neural graphics right inside the engine.
The bottom line? This is not DLSS on your phone, it’s the recipe for DLSS on your phone. Whether or not you’ll see that happening anytime soon depends completely on the speed with which Arm’s licensees put their intellectual property into silicon.
Also read: We saw a gap: Logitech’s Valeriya Kostyukovskaya on the problem MX Keypad solves for developers